Background

This document summarizes the analysis was done on the correlation between historical floods and monthly precipitation in South Sudan.
This analysis was done as exploration for a possible anticipatory action framework on floodings in South Sudan. The reason for looking at monthly precipitation instead of hydrological forecasting models, is that forecasts on monthly precipitation have a significantly longer lead time. This is a very rough first exploration, and doesn’t fully answer the question whether monthly precipitation can be used as an indicator for flooding. However, from the first results the correlations don’t look too promising.
Also note that for this work observed monthly precipitation was used, while for anticipatory action forecasted monthly precipitation should be used.

Definitions

  • The rainy season is defined as being from July till October
  • The spatial level being analyzed is admin1
  • The original format of the flooding and precipitation data is raster data. This data is aggregated to one number per admin1 by taking the mean value of all cells within the given admin.

Historical floodings

To determine historical floodings, we use a dataset by Floodscan. The company determines floodings based on satellite imagery. This data is not openly available.

The variable is the percentage of the area that was flooded. The graph below shows the historical values per admin1. As can be seen the average flooded fraction across the year differs significantly per admin1 region. One reason for this can be constant wet bodies such as lakes (in future analysis these should be filtered out).
More interestingly, we can see clear differences of the maximum flooded fraction per rainy season. These possible indicate severe floods. As can be seen some regions do experience large peaks during the rainy season, such as Jonglei, Lakes, Unity and Warrap. Others see less difference between the rainy and dry season, such as in the Western Bahr el Ghazal and Western Equatoria.

Mean flooded fraction per admin1 in South Sudan from 2000 till 2020

Mean flooded fraction per admin1 in South Sudan from 2000 till 2020

Historical precipitation

The indicator we would like to know the correlation of floods to is monthly precipitation. We firstly only look at observed rainfall, for which we use the CHIRPS dataset. The anomalies in rainfall, compared to the 20 year average of the given month and admin1, are shown below. We can see that there are clear occurrences of more heavy rainfall, indicated by higher green bars. This positive anomaly often co-occurs across several regions, for example in 2001. The magnitude of the anomaly does differ significantly per region.

Monthly precipitation anomaly in South Sudan from 2000 till 2020

Monthly precipitation anomaly in South Sudan from 2000 till 2020

Correlation of historical flooding and precipitation

Since this is a very exploratory analysis, a simple correlation analysis was done. By only looking at the scatter plots and the Pearson correlation. As the results show the correlation is not very strong and from this it could be concluded that monthly precipitation might not be a good indicator for a flood occurring. However, there are two important steps to be taken before concluding this:

  • The correlation should be understood better per region. One example is given for Jonglei where we can already see a completely different pattern.
  • The correlation should be looked at only for the events of flooding, instead of all dates during the rainy season.
Scatter plot of flooded fraction and monthly precipitation

Scatter plot of flooded fraction and monthly precipitation

Correlation matrix of flooded fraction and monthly precipitation, with different aggregation methods

Correlation matrix of flooded fraction and monthly precipitation, with different aggregation methods

Correlation matrix of flooded fraction and monthly precipitation, for the Jonglei region

Correlation matrix of flooded fraction and monthly precipitation, for the Jonglei region